The New England Wireless & Steam Museum is having its annual steam-up on Saturday, October 7. I’m hoping to make it this year, which I haven’t managed to do previously.
Related posts:
Machine tools and glassmaking – visiting the American Precision Museum and the Simon Pearce glassmaking facility.
A retrotech adventure – the Essex Steam Train and Riverboat.
Retrotech, revitalized – a triple-expansion steam engine, built for municipal water pumping, has been restored to operating condition.
Retrotech event, west coast edition – Bill Brandt visits a lumber mill from 1914.
The American Precision Museum is wonderful!
Yes, great place. It would be nice if they could schedule some of the demos in advance.
Retronukes?
Palisades nuclear power plant lookin’ to become the first recommissioned nuke in U.S. history
https://hotair.com/tree-hugging-sister/2023/09/13/palisades-nuclear-power-plant-lookin-to-become-the-first-recommissioned-nuke-in-u-s-history-n577807
Not exactly retro tech but still tech. Some may find it a little tedious, what can I say, I’m easily amused. It shows some of the process of producing a fairly complicated part on a five axis turn mill. Notice just how far the idea that you can download a drawing and push the start button is from reality. A lot of the discussion revolves around just general machining, feeds and speeds, still the basics of separating what you don’t want from what you do.
https://www.youtube.com/watch?v=O8WYTBoHNeE
Software aids are already used extensively. I’m sure there are numerous, numerous AI projects in this area. One of the things I’ll be looking for in judging progress is when the people with six and seven figure machines, running four and five figure tools on workplaces worth that and more start using AI. In this context, a crash can shake the whole building. As a spectator rather than a player, I’m sure it’s coming, there are already a lot of tools and probably already comes down to a matter of definitions. I’ve already said I think the opacity of large language models is a, possibly, insurmountable problem.
A little more OT. If you’re in Dallas and looking for an interesting few hours:
https://cavflight.org/
Would you really want to use a LLM for machining planning, anyhow? Isn’t the nature of the problem that your AI advisor will need some pretty structured and mostly numerical information…tolerances, materials, cost vs time tradeoffs, etc?
The part specification basically establishes two complex surfaces which the surface of acceptable parts need to fall between. At a simple level, the diameter of a round feature needs to be between a minimum and a maximum diameter. With cad, it’s possible to establish that envelope with far higher precision that it is possible to measure and measuring to confirm conformance is another whole problem. There are also constraints beyond simple geometry such as surface roughness or roundness that have to be met. All of this is set in stone but it’s only the beginning of the process of making a part. MAking the part reliably and economically is the rub.
The number of degrees of freedom involved in making a part like the one in the video makes computer chess look simple. Only 32 pieces and 64 squares. I wouldn’t know how to start to count all the possible sequences that would result in an acceptable part. This is the combinatorial explosion that has stymied all the predictions of immanent human obsolescence. Possibly, the most important human ability is making a plausible start on such problems while computers churn endlessly seeking the “optimum”. Think of it like finding a square root by Newton’s method, even starting with a wildly wrong guess will add, at most, about two iterations.
Going back to the video, there are several places where the programmer changes parameters, usually by reducing the load in pursuit of a more predictable outcome, less chance of tool breakage or better finish. All, undoubtedly, based on experience, some of it pretty expensive I’d bet. This sort of judgement is also where computers seem to be limited.
There are lots of different aids that can be used. CAM software has long allowed tool paths to be programmed at an abstract level above G Code but again, just like the computer programmer using snippets of assembly language, you see naked G code has to be added from time to time by hand. There are a lot of tools focused on simulation and finding gross errors that work like spell checkers. Smart people use all the help they can get.
I just had a part made. I provided a solid model and a paper drawing. I had the luxury of talking with the person that programmed and ran the part to clear up everything, and there’s always something in my experience. Yet there was still a problem. The interaction of the material and the tool produced a hole that was ever so slightly undersized. This was easily fixed, there were only five pieces, so reaming the holes wasn’t a problem for more than few minutes, but it shows that nothing is as simple as it looks. On the other hand, the solid model allowed all the other tricky geometry to be communicated without the sort of intricate drawings that would have been necessary otherwise.